Research On Short Term Traffic State Prediction Method Based On Sequence Model

With the rapid growth of the economy,the development momentum of the automotive industry is strong,China’s transportation has been confronted with immense difficulties,thus making the issue of traffic congestion more evident.It is more appropriate to describe the rush hour as a difficult time for a car to move at an inch.The current situation of transportation in China includes poor transportation facilities,traffic congestion,low road operation efficiency,poor driving safety,serious pollution caused by exhaust emissions,and high energy consumption.For this reason,Intelligent Transportation Systems(ITS)have become a panacea for solving this problem,and have become recognized as a hot spot for solving road traffic congestion,improving road operation efficiency,improving air quality,and reducing environmental problems.The way for intelligent transportation to achieve this result is to accurately predict shortterm traffic flow.To maximize the value of the vehicle data,and to enhance the precision of state recognition and prediction,it is imperative to investigate and examine novel state prediction techniques to acquire all the wealth information in traffic data,and to augment the accuracy and dependability of state recognition and prediction.This article uses Mn DOT RTMC traffic flow data from the Twin Cities metro area in Minnesota,USA to make short-term predictions of urban road traffic flow.The research process of the article is as follows:(1)The importance of traffic flow in forecasting was discussed,and the relevant model theory of deep learning was introduced,along with the pertinent concepts and parameters of traffic flow.(2)Preprocess the Mn DOT RTMC traffic flow data in the Twin Cities metro area of Minnesota,USA,and repair and denoise the lost and collected abnormal data.Analyze the characteristics of traffic flow from several aspects such as nonlinearity,periodicity,uncertainty,and time correlation.Quantify the correlation of traffic data for a week using covariance and Pearson correlation coefficient methods,Provide support for predictions.(3)Using MATLAB software to analyze the data,using the Gap Statistical algorithm to determine the value of K.By applying the K-means++algorithm,each group of 15 minute data is effectively grouped and clustered.The Davies Boldin index determines the number of clusters as 6.Then,a traffic state prediction model is established based on time series.By analyzing RNN,LSTM Three models,including Transformer,were used to determine the corresponding input and output of the traffic state model,as well as the varying degrees of errors and accuracy statistics of several models.The final model structure of the three models was determined by adjusting parameters such as sample size batch size、learning rate(lr)、and iteration times epochs to 100、512、and 0.0001.The model performance was higher when the training set was input into the model for training,It was found that the accuracy of the three models on the training set exceeded 75%,and the accuracy on the validation set was between 63%-70%,providing strong support for the proposal of traffic management measures and providing reference for future research.